Support np arrays for Vertical Federated Learning

[edumugi]

Experimenting with the federated plugin (#8424) in a vertical environment I noticed that giving data as a numpy array to the DMatrix is not supported yet.

It should be a really easy fix since we only need to support the data_split_mode arg in the _from_numpy_array function.

Steps would be:

• Add new arg data_split_mode to _from_numpy_array in file python-package/xgboost/data.py
• Read data_split_mode from config in function XGDMatrixCreateFromDense inside file src/c_api/c_api.cc

Should I open a PR for these changes?

[trivialfis]

The federated learning support is still working in progress. cc @rongou .

[rongou]

Yes, feel free to send @trivialfis and me a PR.

[trivialfis]

Thank you for the feature!